黄河三角洲湿地是我国北方典型的滨海湿地，蕴藏着极其丰富的生物资源、土地资源、油气资源以及旅游资源等，也是重要的鸟类迁徙中转站、越冬栖息地和繁殖地。近年来，由于人为干扰以及自然因素的共同作用，使得当地的生态环境脆弱性日益凸显。因此，开展黄河三角洲湿地生态环境脆弱性评估，识别造成研究区脆弱性的主要驱动因子，具有重要的科学意义。
  本文针对黄河三角洲生态系统复杂性和特异性，基于贝叶斯网络模型的方法进行生态环境脆弱性评估和预测。首先，在总结前人研究的基础上，筛选了代表研究区生态环境特征的指标，构建了包含8个因子的准则层、11个因子的指标层的脆弱性评估体系。其次，构建了基于贝叶斯网络的黄河三角洲生态环境脆弱性预测模型，利用交叉验证的方法进行模型筛选。在此基础之上，对黄河三角洲生态环境现状进行了评估，系统分析了研究区的脆弱性分布及原因，从多角度探讨指标与脆弱性之间的关系。模型敏感性分析表明：对研究区脆弱性影响较大的三个指标分别为：NDVI、土地利用和植被类型。最后，利用构建的贝叶斯网络模型，设置了与人类活动、气候、NDVI等相关情景，预测不同情景下的脆弱性状态，并提出了相应的管理建议，为黄河三角洲资源开发利用与生态环境保护提供了一定的科学依据。
  分析表明，研究区以轻微脆弱和轻度脆弱为主，约占总面积2/3，中度脆弱约占2/15，重度脆弱和极度脆弱约占总面积的1/5。通过模型性能分析，筛选一个错误率仅为17.35%的优良模型，为情景分析奠定了基础。设置的一系列情景分析表明，基于贝叶斯网络以不确定知识表达形式进行生态脆弱性预测是一种有效的方法。综上，贝叶斯网络模型可以综合各类生态环境信息，并可以有效融合，从而推断各因子之间的因果关系，为湿地生态环境脆弱性的预测提供了一种新的思路。

关键词：黄河三角洲，贝叶斯网络，脆弱性，生态环境，预测
 

   The Yellow River delta (YRD) is a typical coastal wetland in northern China with rich natural resources, unique habitats and landscapes. On the other hand, it is the important relay stations, wintering habitat and breeding place for migrating birds in eastern Asia. In recent years, the eco-environment in Yellow River delta became more and more vulnerable due to intensive human innervations and natural variability. Therefore, it is necessary and urgent to fully assess the ecological and environmental vulnerability and clearly identifying the driving factors.
   In this thesis, a Bayesian network (BN) model was constructed and utilized to assess and predict the ecological and environmental vulnerability in YRD. Firstly, based on previous studies about ecological and environmental vulnerability in this area,an assessment framework has been proposed including 8 factors criteria layerindices and 11 factors index level indices, where all indices were carefully screened and selected. Secondly, a Bayesian network model was constructed based on the index system. BN model is an uncertain knowledge representation model based on probability theory and graph theory, which plays an important role in uncertain reasoning. Cross validation was also applied to select the optimal model, with which both the current status of vulnerability was assessed. Specifically, the spatial distribution and causes of ecological and environmental vulnerability was reasoned using BN model. Moreover, the sensitivity analysis showed that NDVI, land use and vegetation types were the three primary factors that determine ecological and environmental vulnerability significantly. Finally, different scenarios about human activities, climate change and NDVI change were set and predict the evaluation of ecological and environmental vulnerability in YRD by considering the local economy and social developments so that suggestions were proposed to solve the ecological and environmental problems.
   The analysis of the current situation shows that the study area is dominated by slight vulnerability and mild vulnerability, which accounts for about 2/3 of the total area, and the moderate vulnerability is about 2/15, and the severe vulnerability and extreme vulnerability are about 1/5 of the total area. Through the performance analysis of the model, a good model with an error rate of only 17.35% is selected, which lays the foundation for the scenario analysis. A series of scenario analysis shows that the uncertain knowledge representation of ecological vulnerability prediction is an effective method based on Bayesian network, which provides a new idea for the prediction of ecological vulnerability in the Yellow River delta. In summary, the Bayesian network model can integrate various ecological environment information, and it can effectively fusion these information, besides it can infer a causal relationship between each factor, which provides a new idea for wetland ecological environment vulnerability prediction.
Keywords: Yellow River Delta, Bayesian network, vulnerability, ecological environment, prediction